question_answer_matching/data_utils.py at master · sheep-for/question_answer_matching · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
# encoding:utf-8

import numpy as np
import codecs
import re
import os
import pandas as pd
import collections
from operator import itemgetter
from itertools import chain
from functools import reduce
import random
import gensim.models.word2vec as word2vec


import platform
if platform.system() == 'Windows':
    import pickle
else:
    import cPickle as pickle


def build_pre_train_emb_matrix(emb_vec_file, vocab_file):
    embedding = []
    temp_emb = dict()
    question_vec_file = codecs.open(emb_vec_file, 'r', 'utf-8')
    for index, line in enumerate(question_vec_file.readlines()):
        if index != 0:
            key = line.strip().split(' ')[0]
            vector = list(line.strip().split(' ')[1:])
            temp_emb[key] = vector
    question_vec_file.close()

    vocab_word = list(chain.from_iterable(pd.DataFrame(
        pd.read_csv(vocab_file, encoding='utf_8_sig'), columns=['word']).values))
    count = 0

    for i, item in enumerate(vocab_word):
        if item in temp_emb:
            vec = temp_emb[item]
            embedding.append(vec)
            count += 1
        else:
            print("new word.", item, "idx ", i, " initialized as <unk>")
            vec = temp_emb['\0']
            embedding.append(vec)

    del temp_emb
    return np.array(embedding, np.float32)


def get_keyword_num_log_tfidf(sentence_length, input_type="question"):
    # for q: keyword_num = min(10 * ln(x), x)
    # for a: keyword_num = min(x[logx] / lg(2x), x)
    """
    :param sentence_length: numpy.int32, sequence_length of inputs
    :param input_type: question or answer
    :return: the same shape and type with sentence_length,
    the keyword_num for attention_layer and representation_layer
    """

    sentence_length = np.array(sentence_length, np.int32)
    if input_type == "question":
        keyword_num = np.minimum(
            np.int32(10 * np.log(sentence_length)), sentence_length)
    elif input_type == "answer":
        keyword_num = np.minimum(
            np.int32(sentence_length * np.int32(np.log10(sentence_length)) / np.log10(2.0 * sentence_length)),
            sentence_length)
    else:
        keyword_num = sentence_length
    if 0 in keyword_num:
        keyword_num = np.where(keyword_num > 0, keyword_num, np.ones_like(sentence_length))
    return keyword_num


class TrainingDatasetLoader(object):

    def __init__(self, num_answer_for_per_question, batch_size, train_file):
        # self.num_answer_for_per_question = num_answer_for_per_question
        self.question_batch_size = batch_size * num_answer_for_per_question // 8
        self.train_file = train_file
        self.question = []
        self.question_length = []
        self.answer = []
        self.answer_length = []
        self.pointer = 0

        self.create_batches()

    def create_batches(self):
        question_input = codecs.open(self.train_file, 'r', 'utf-8')
        question_data = question_input.readlines()
        for index_q, line_q in enumerate(question_data):
            line_str = line_q.strip().split('\t')
            line_int = [int(x) for x in line_str[0].split()]
            self.question.append(line_int)
            self.question_length.append(len(line_int))
            for i in range(1, 6):
                answer_i_str = line_str[i].split()
                answer_i_int = [int(x) for x in answer_i_str]
                self.answer.append(answer_i_int)
                self.answer_length.append(len(answer_i_int))
        del question_data
        question_input.close()

        self.num_batch = len(self.question_length) // self.question_batch_size
        self.question = self.question[: self.num_batch * self.question_batch_size]
        self.question_length = self.question_length[: self.num_batch * self.question_batch_size]
        self.answer = self.answer[: self.num_batch * 5 * self.question_batch_size]
        self.answer_length = self.answer_length[: self.num_batch * 5 * self.question_batch_size]

        self.point = list(range(self.num_batch))
        random.shuffle(self.point)

    def next_batch(self):
        question_batch_length_before_tile = self.question_length[
            self.point[self.pointer] * self.question_batch_size: (self.point[self.pointer] + 1) * self.question_batch_size]
        max_question_batch_length = max(question_batch_length_before_tile)
        question_batch_before_tile = self.question[
                                     self.point[self.pointer] * self.question_batch_size: (self.point[self.pointer] + 1) * self.question_batch_size]
        for idx_q, q in enumerate(question_batch_before_tile):
            if len(q) < max_question_batch_length:
                question_batch_before_tile[idx_q] = q + [0] * (max_question_batch_length - len(q))
        question_batch = []
        question_batch_length = []
        for question_before_tile in question_batch_before_tile:
            for _ in range(8):
                question_batch.append(question_before_tile)
        for question_length_before_tile in question_batch_length_before_tile:
            for _ in range(8):
                question_batch_length.append(question_length_before_tile)
        del question_batch_before_tile
        del question_batch_length_before_tile

        answer_batch_length_before_tile = self.answer_length[
                                          self.point[self.pointer] * 5 * self.question_batch_size: (self.point[self.pointer] + 1) * 5 * self.question_batch_size]
        max_answer_batch_length = max(answer_batch_length_before_tile)
        answer_batch_before_tile = self.answer[self.point[self.pointer] * 5 * self.question_batch_size: (self.point[self.pointer] + 1) * 5 * self.question_batch_size]
        for idx_a, a in enumerate(answer_batch_before_tile):
            if len(a) < max_answer_batch_length:
                answer_batch_before_tile[idx_a] = a + [0] * (max_answer_batch_length - len(a))
        answer_batch = []
        answer_batch_length = []
        for i in range(self.question_batch_size):
            for j in range(4):
                answer_batch.append(answer_batch_before_tile[(5*i+4)])
                answer_batch_length.append(answer_batch_length_before_tile[(5*i+4)])
                answer_batch.append(answer_batch_before_tile[(5*i+j)])
                answer_batch_length.append(answer_batch_length_before_tile[(5*i+j)])
        question_keyword_num = get_keyword_num_log_tfidf(question_batch_length, "question")
        answer_keyword_num = get_keyword_num_log_tfidf(answer_batch_length, "answer")

        self.pointer = (self.pointer + 1) % self.num_batch
        if self.pointer == 0:
            random.shuffle(self.point)
            print('data shuffle. point: ', self.point[:5])
        return np.array(question_batch, np.int32), np.array(question_batch_length, np.int32), question_keyword_num, np.array(
            answer_batch, np.int32), np.array(answer_batch_length, np.int32), answer_keyword_num

    def reset_pointer(self):
        self.pointer = 0
        random.shuffle(self.point)


class TestingDatasetLoader(object):

    def __init__(self, num_answer_for_per_question, batch_size, test_file):
        # self.num_answer_for_per_question = num_answer_for_per_question
        self.question_batch_size = batch_size * num_answer_for_per_question // 5  # 160/5=32
        self.test_file = test_file
        self.question = []
        self.question_length = []
        self.answer = []
        self.answer_length = []
        self.pointer = 0

        self.create_batches()

    def create_batches(self):
        question_input = codecs.open(self.test_file, 'r', 'utf-8')
        question_data = question_input.readlines()
        for index_q, line_q in enumerate(question_data):
            line_str = line_q.strip().split('\t')
            line_int = [int(x) for x in line_str[0].split()]
            self.question.append(line_int)
            self.question_length.append(len(line_int))
            for i in range(1, 6):
                answer_i_str = line_str[i].split()
                answer_i_int = [int(x) for x in answer_i_str]
                self.answer.append(answer_i_int)
                self.answer_length.append(len(answer_i_int))
        del question_data
        question_input.close()

        self.num_batch = len(self.question_length) // self.question_batch_size
        self.question = self.question[: self.num_batch * self.question_batch_size]
        self.question_length = self.question_length[: self.num_batch * self.question_batch_size]
        self.answer = self.answer[: self.num_batch * 5 * self.question_batch_size]
        self.answer_length = self.answer_length[: self.num_batch * 5 * self.question_batch_size]

        self.point = list(range(self.num_batch))

    def next_batch(self):
        question_batch_length_before_tile = self.question_length[
            self.point[self.pointer] * self.question_batch_size: (self.point[self.pointer] + 1) * self.question_batch_size]
        max_question_batch_length = max(question_batch_length_before_tile)
        question_batch_before_tile = self.question[
                                     self.point[self.pointer] * self.question_batch_size: (self.point[self.pointer] + 1) * self.question_batch_size]
        for idx_q, q in enumerate(question_batch_before_tile):
            if len(q) < max_question_batch_length:
                question_batch_before_tile[idx_q] = q + [0] * (max_question_batch_length - len(q))
        question_batch = []
        question_batch_length = []
        for question_before_tile in question_batch_before_tile:
            for _ in range(5):
                question_batch.append(question_before_tile)
        for question_length_before_tile in question_batch_length_before_tile:
            for _ in range(5):
                question_batch_length.append(question_length_before_tile)
        del question_batch_before_tile
        del question_batch_length_before_tile

        answer_batch_length = self.answer_length[
                                          self.point[self.pointer] * 5 * self.question_batch_size: (self.point[self.pointer] + 1) * 5 * self.question_batch_size]
        max_answer_batch_length = max(answer_batch_length)
        answer_batch = self.answer[self.point[self.pointer] * 5 * self.question_batch_size: (self.point[self.pointer] + 1) * 5 * self.question_batch_size]
        for idx_a, a in enumerate(answer_batch):
            if len(a) < max_answer_batch_length:
                answer_batch[idx_a] = a + [0] * (max_answer_batch_length - len(a))
        question_keyword_num = get_keyword_num_log_tfidf(question_batch_length, "question")
        answer_keyword_num = get_keyword_num_log_tfidf(answer_batch_length, "answer")

        self.pointer = (self.pointer + 1) % self.num_batch

        return np.array(question_batch, np.int32), np.array(question_batch_length, np.int32), question_keyword_num, np.array(
            answer_batch, np.int32), np.array(answer_batch_length, np.int32), answer_keyword_num

    def reset_pointer(self):
        self.pointer = 0